A Review of Multi Agent Approaches for Inventory Control in Supply Chain: Future Prospectus

A Review of Multi Agent Approaches for Inventory Control in Supply Chain: Future Prospectus Peeyush Vats Poornima College of Engineering ABSTRACT Inventory control decisions are always very critical issues in an effective supply chain. One of the major decision areas in supply chain management is inventory. There are mainly two classical approaches for inventory control i.e. deterministic approach, probabilistic approach or stochastic approach. In this paper another approach i.e. multi agent approach is discussed for inventory control in supply chain. Many researchers applied Multi agent modeling system in various areas of supply chain and inventory and found better results as compared to conventional methods. In this paper a review of multi-agent modeling approaches for inventory control is discussed. In this paper a new inventory control system ICMAS is also introduced. Keywords Inventory control system, Multi Agent, Review, INTRODUCTION Generally inventory can be defined as the list of items held in a stock and the inventory management is the functions responsible for all decisions about stock in an organization. It makes decision for policies, activities and procedure to make sure the right amount of each item held in a stock at any time [1].Inventory decisions are very critical from day to day operation of an organization and it is very difficult to keep customer satisfaction level high. An operational inventory decision depends on optimal stock level at each location as the market demand fluctuates. Inventory Control policies must be overlooked to determine correct level of supply and demand at order and reorder points. Generally the holding cost of inventory lies almost in between 20 percent to 40 percent of thetotal cost of inventory. The ultimate goal in inventory control policies is to reduce the inventory costs and to keep the customer satisfaction level high. IMPORTANCE OF INVENTORY IN AN EFFECTIVE SUPPLY CHAIN Inventory exists in supply chain because of mismatch between demand and supply. Inventory can be used to increase the amount of demand in supply chain by having the product ready and available when the customer wants it. Another significant role of inventory in supply chain is that to reduce the cost by exploiting economies of scale that may exists during production and distribution. Inventory impacts the assets held, the cost incurred, and the responsiveness provided in the supply chain[2]. MODELING APPROACHES IN INVENTORY CONTROL It is very essential to maintain inventory levels for different kinds of organizations like manufacturing units, retailers, factories and enterprises. A better inventory management would constantly develop organizational productivity, decrease costs, and contribute to responsible use of scarce capital. Other than raw material inventory, other forms of inventory include work in-process inventories, supply inventories, component inventories, and finished goods inventories. The most important aim of inventory management is to decide how much resources or inputs are to be arranged and when to order so as to reduce production cost, while conforming to the essential requirements. Due to ranging abnormality of the production inventory, no specific 512 Peeyush Vats

inventory model has general relevance to the whole variant inventory situations.as a result, a range of inventory models have appeared which address specific inventory problems. The classic inventory model is generally used either to forecast optimum inventory or to evaluate two or more inventory systems. Inventory modeling deals with determining the level of a commodity that a business must maintain to ensure smooth operation. The basis for the decision is a model that balances the cost of capital result from holding too much inventory against the penalty cost resulting from inventory shortage. The principal factor affecting the solution is the nature of the demand: Deterministic or Probabilistic. In real life, demand is usually probabilistic, but in some cases the simpler deterministic approximation may be acceptable [4]. Generally there are two methods for modeling the inventory i.e. deterministic and probabilistic or stochastic. 4.1 Deterministic approach: The deterministic approach concedes a single best estimation of inventory reserves grounded on recognized engineering, geological, and economic information. Deterministic models of inventory control are used to determine the optimal inventory of a single item when demand is mostly largely obscure. Under this approach inventory is built up at a constant rate to meet a determined, or accepted, demand. 4.2 Probabilistic approach: The probabilistic approach employs the known economical, geological and engineering data to produce a collection of approximate stock reserve quantities and their related probabilities. Each inventory reserve categorization gives a signal of the prospect of revival. The advantage of a probabilistic approach lies in the fact that by using values lying within a bandwidth and modeled by a defined distribution density, the reality can be modeled better than by using deterministic figures. 4.3 Stochastic approach: Stochastic approach can also be used for inventory control. Such approaches are used when demand is uncertain. Stochastic approaches are more relevant and more realistic as compared to deterministic and probabilistic approaches. Since they regard the cost of arranging, cost of shortfalls and the cost of stacking away, and try to attempt to formulate an optimal inventory plan. MULTI AGENT MODELING APPROACH FOR INVENTORY CONTROL Conventional inventory management is becoming obsolete due to rise in the global sourcing, contract manufacturing and dynamic nature of inventory. So we have adopted multi agent modeling approaches for inventory control problems.agent based models (ABMs) or multi agent system (MASs) consists of set of elements (agents) characteriz ed by some attributes, which interact each other through the definition of appropriate rules in a given environments[5]. Through ABMs, it is possible implementing an environment with its feature, forecasting and exploring its future scenarios experimenting possible alternative decisions, setting different values for decision variables and analyzing the effects of these changes. According to the definition of Wooldridge and Jennings (1995) [6], an agent is a computational system interacting with the environment that can be endowed with the following features: Independence: Each agent acts without the direct control of human beings or other devices Social Abilities: Interactions occur among entities through a communication language in order to satisfy the objectives. Re-activeness: Agents answer in a precise way to signals coming from the environment. Pro-activeness: Agents are endowed with goal-directed behaviors. They take the initiative in order to satisfy their objective. In order to pursue their own objectives, agents have to interact and communicate. Communication capabilities include the abilities to receive and send messages. This is necessary to ensure a coordination mechanism among agent themselves, in order to prevent and avoid conflicts among agents' objectives for accessing resources or achieving objective. Implementing a cooperation-based coordination mechanism means employing planning approaches to reduce resource contention and ensure the achievement of global objectives. 513 Peeyush Vats

6. Categorization of multi-agent system architecture:this approach can be distinguished in to two main categories[7] Distributed approaches: In this approach agents are endowed with self-organizing rules for resource sharing and goal purchasing Centralized approaches: In this approach a mediator agent is assigned with task of regulating and supervise agents behaviors MULTI AGENT SYSTEM CENTRALIZED DECENTRALIZED Figure 1: Multi agent system categories MULTI-AGENT APPROACHES VS. CLASSICAL APPROACHES According to the classical approach of operations research, an optimization problem tries for searching the best solution, according to a given criterion, among a set of feasible solutions. Optimization algorithms are generally step-by-step procedures for solving optimization problems. In other words an optimization algorithm can be applied to any instance of that problem to produce a feasible solution. Optimization algorithms may be exact if they canprovide the optimal solution. Optimization algorithms may be heuristic if they cangive a good solution but not necessarily the optimal [8]. Heuristics are particularly useful to solve complex problems i.e. problems belonging to the class of NP-hard problems. For a description of the most recent trends in the field of heuristic methods see Burke and Kendall (2005) [9]. Due to their characteristics, ABMs have been recently using as a promising heuristic techniques to solve problems whose domains are distributed, complex and heterogeneous. Madejski (2007) [10]pointed out that, Agent Based Mod eling for optimization objectives can be designed according to a physical or a functional decomposition scheme. In the first case, agents represent physical entities (like, workers, machine tools, resources, vehicles) involved in the specific probl em to be solved and in the second case, agents are assigned to some functions (for instance, scheduling, sequencing, material handling) based on some rules aiming at reproducing optimizing behaviors. Davidsson, Holmgren, and Persson (2007) [11] proposed a theoretical framework to compare the characteristics of agent-based and of classical heuristic approaches, based on a set of dimensions such as size, modularity, changeability/time scale, solution quality, computational complexity. Considering the size and the modularity of a problem, agent-based approach supports the partition of the global problem into a number of smaller local problems. A large-sized optimization problem could be converted well in to a smaller problem which presents the characteristics of modularity. This way, reducing the size of singles sub-problems could help in the achievement of better solutions. Since agents are capable to continuously monitor the state of the environment and typically do not have to make very complex decisions, they could be able to respond to changes quickly. In particular, if the problem is characterized by a high dynamicity ( time scale/changeability), ABMs can present higher reactivates degrees, which could permit to found appropriate solutions. On the other hand, optimization techniques often require a relatively long time to respond to changes in variables and parameters of the problem, as they often need a complete restart. From the point of view of the quality of solution, since agent based approaches are distributed, they do not have a global view of the state of the system, which is often necessary in order to find a truly good solution. Therefore, the quality of the solution provided by a classical heuristic could be better. However in some problems which fit the characteristics of ABMs, their application can produce competitive results. In terms of computational times, agent-based approaches can provide some advantages thanks to their ability to divide problems in sub- 514 Peeyush Vats

problems. However, computational advantages can be offset by the need for frequent interaction in order to coordinate activities according to a given paradigm, as explained in the previous section; classical approaches present higher computational complexity, but no communication costs, as they are characterized by high centralization[5]. On the basis of these considerations, ABM could be an interesting approach when the size of the problem is large; the domain is modular in nature and the structure of the domain changes frequently. In particular the most of the applications are in manufacturing process due to the complexity of the search space with a high number of decision variables, parameters and constraints (Shen, Wang, &Hao, 2006) [12], the high variability of the domain which requires an ability to react to short-term disturbances (Kornienko, Kornienko, &Priese, 2004) [13]. Indeed, in manufacturing problems, the domain changes very frequently, and agent-based approaches can offer good features to deal with this situation. The potential of agent technology is also suitable for modeling practical supply chain management problems, where analytic or optimization results cannot be easily applied or global information sharing and central coordination are difficult to achieve (Ahn& Lee, 2004) [14]. From certain point of views ABMs and classical heuristics have complementary characteristics. For this reason there is an increasing interest towards approaches embedding optimization techniques within an ABM schema. In practice the integration can be performed through two fundamental ways: Utilizing an optimization technique for strategic planning and ABMs for operational and tactic replanning, i.e., for performing local adjustments of the initial plan. Embedding optimizing behaviors in physical agents, by translating search algorithms into agents behavior. APPLICATION OF MAS IN INVENTORY: A REVIEW OF APPLICATION OF MAS IN INVENTORY Dejun Chen et al (2008) worked [15]on the inventory scheduling model based on agent orient petri net supply chain. According to them it is very useful tool of analysis for the modeling of inventory scheduling in supply chain system.qingqi Long and Wenyu Zhang (2013) [16]developed an integrated frame work for agent based inventory production and transportation modeling and distributed simulations of supply chain. This paper s multi-level framework comprises four levels from domain modeling to the implementation of multi-agent systems and integrates the agent-based modeling and distributed simulation theory, a four-layered conceptual agent modeling framework, a meta-agent class library, and a multi-agent based distributed simulation platform to build an agent-based inventory production transportation model and simulate it in a distributed way. It extends the conceptual modeling framework. In their PaperChengzhi Jiang and Zhaohan Sheng (2009) [17] proposed a case based learning algorithm (CRL) for dynamic inventory control in a multi agent supply chain system. They also considered a framework for general learning method based on proposed one, which may be helpful in all aspects of supply-chain management (SCM).Chang Ouk Kim et al (2010) [18] considered a multi -stage inventory control problem with non-stationary customer demand under a customer service-level constraint. They proposed a multi-agent based model for distributed inventory control systems. In this model, the agent at the first stage is called a retail agent and those at the remaining stages are called supply agents. DONG Fu-gui et al (2012) [19] proposed an agent based simulation model of a single point inventory system. Through simulation by multiagent system by any logic software they compared the two continuous replenishment strategies, the (R, S) and (Q, R) strategies and proved that (R, S) strategy is better than (Q, R) strategy by simulation method. A PROPOSED MODEL FOR INVENTORY CONTROL BY MULTI-AGENT SYSTEM (ICMAS) There isno specific inventory model which covers the complete variant inventory situations. A specific inventory modeladdresses to a specific inventory problems.as a result, a very wide range of inventory models have been proposed. Hence such type of a model is required which covers almost all the aspects of inventory 515 Peeyush Vats

problems. In the classical approach it is not possible to cover all the aspects of inventory problem. Hence therefore a multi agent model is proposed for the controlling the inventory problems. In the multi agent approach all the aspects may be covered for optimizing an inventory problem. This model suggests various types of agents like incoming message agent or receiver agent, supplier agent, warehouse agent, price agent, location agent, market agent, product agent, finance agent and co-coordinator agent. Each agent has its own specific features and functions and can communicate any one of the above agents.in a business firm each day begins with inventory checking, order handling, reordering and random demand producing. Any demand by a customer will trigger a series of events and make agent's properties or behaviors change [19]. A receiver agent receives all the information according to balancing demand and supply. This receiver agent now transmits the information to all inventory agents like supplier agent, warehouse agent, price agent, location agent, market agent, product agent and finance agent. These agents are also capable to communicate with each other and coordinate with a coordinator agent. This coordinator agent tries to optimize the inventory problem after getting information from all of the above agents. Figure 2: A proposed model for inventory control by multi-agent system (ICMAS) CONCLUSION In this paper different approaches used in inventory control were discussed. Generally there were two approaches like deterministic approach and probabilistic approach for inventory control. Some researchers applied Multi agent approaches in the inventory control system. On the basis of literature available, it can be said that Multi agent modeling is better than convention approach and provides better simulation results. Multi Agent System (MAS) may be used as a decision support system. 516 Peeyush Vats

References: [1]. Donald Waters (2003), "Inventory Control and Man agement", John Wiley & Sons Ltd, 2 nd Edition, pp 4, 7, ISBN 0-470-85876-1. [2]. Sunil Chopra, Peter Meindl and D. V. Kalra (2013), "Supply Chain Management, Strategy, Planning and Operation", Dorling Kindersley (India) Pvt Ltd., 5 th Edition, pp 54-55. ISBN 978-81-317-8209-9. [3]. S. D. Sharma (2010),"Operations Research" Inventory/Production Management -I, KedarnathRamnath Publications, Fifteenth Edition, pp 665-666 [4]. Hamdy A. Taha (2008), "Operations Research: An Introduction", Pearson Edu cation (Singapore) Private Limited, 8 th Edition, pp 431-449, 559-573, ISBN 81-7807-757-X. [5]. M. Barbati, G. Bruno and Genovese (2012), "Application of Agent-Based Models for Optimization Problems: A Literature review", Expert system with Applications, 39(2012), pp 6020-6028. [6]. Wooldrige, M., and Jennings, N. (1995), "Intelligent Agents: Theory and Practice", Knowledge Engineering Review, 10(2), pp 115-152. [7]. Georgios Andreas, K.-D. Bouzakis, P. Klazoglou and K. Niwtaki (2014), "Review of Agent -Based system in Manufacturing Section", Universal Journal of Mechanical Engineering 2(2), pp 55-59. [8]. Frederick S. Hiller and Gerald J. Lieberman (2010), "Introduction to Operations Research" McGraw Hill Publication, 9 th Edition, pp 607, ISBN 978-0-07-337629-5 [9]. Bruke E. K. and Kendall G (2005)," Search Methodologies: Introductory Tutorials in Optimization and Decision Support Techniques", Springer, ISBN: 978-0-387-23460-1. [10]. Madejski J. (2007), "Survey of the Agent Based Approach to Intelligent Manufacturing", Journal of Achievements in Materials and Manufacturing Engineering, 21(1), pp 67-70. [11]. Davidsson, P., Henesey L., Ramstedt L., Tornquist J., and Wernstedt, F. (2005). "Agent Based Approaches to Transport Logistics", Transportation Research Part C: EmergingTechnologies, 13(4),pp255 271. [12]. ShenW., Wang L., and Hao, Q. (2006). "Agent -based Distributed Manufacturing Process Planning and Scheduling: A state-of-the-art survey", IEEE Transactions onsystems, Man, and Cybernetics-Part C: Applications and Reviews, 36(4), pp563 577. [13]. Kornienko S., Kornienko O., and Priese, J. (2004). "Application of multi -agent planningto the assignment problem". Computers in Industry", 54, pp273 290. [14]. Ahn, H., and Lee, H. (2004),"Agent Based Dynamic Network for Supply Chain Management", BT Technology Journal, 22(2), pp18 27. [15]. Dejun Chen, Zude Zhou and Rui Hu (2008), "Research on the Inventory Scheduling on Agent-oriented Petri net in Supply Chain", Kybernetes, Emerald group publishing limited, Volume 37, No. 9/10, pp 1234-1241. [16]. Qingqi Long and Wenyu Zhang (2014), "An Integrated Framework for Agent based Inventory -Production- Transportation Modeling and Distributed Simulation of Supply Chain", Journal of Information Sciences, Article in Press, http://dx.doi.org/10.1016/j.ins.2014.02.147. [17]. Chengzhi Jiang and Zhaohan Sheng (2009), "Case-based Reinforcement Learning for Dynamic Inventory in a Multi-Agent Supply-Chain", Expert System with Application, Volume 36, pp 6520-6526. [18]. Chang Ouk Kim, Ick-Hyun Kwon and ChoonjongKwak (2010), "Multi -Agent Based Distributed Inventory control Model", Expert System with Application, Volume 37, pp 5186-5191. [19]. DONG Fu-gui, LIU Hui-mei and LU Bing-de (2012), "Agent -based Simulation Model of Single Point Inventory System" System Engineering Procedia 4 (2012), The 2 nd International Conference on Complexity Science and Information Engineering, pp 298-304. 517 Peeyush Vats